Initially I thought: wow, finally I can play with GPT-3 without going through the closed beta application process. But the tool requires an OpenAI API key. The linked website doesn't specify that with a single word. It's only mentioned on Github.
I've been there myself tbh. When you have worked a lot on something, it's often hard to take a step back and specify all of the basic things that people first exposed to your invention don't know about.
You know about it, so it's definitely not "one of the most closed projects in existence." It is perhaps one of the more exclusive and overly-hyped projects.
I think the word 'open' was meant to imply open source, not to distinguish between covert ai ops and public ai ops. In that case I think they haven't done enough to make it open to the public.
I am looking forward to all the new AI startups that launch when the OpenAI GPT-3 API goes out of beta. Their secret sauce will be nothing more than a handcrafted prompt, which would be like launching a company based on a fancy SQL query used to query a 3rd party database.
There are multiple groups working on reimplementations of GPT-3. It's not that difficult, just requires a certain level of compute resources which a lot of BigCos and large organizations have on tap already.
I'm more curious to see what happens when single developers are able to launch products which are vastly better than existing solutions, using just a couple days of effort.
1) Big Corp approaches the single dev saying that Famous Boss of Big Corp is impressed by his project and would like to hear more about it and maybe acquire his project
2) Single dev shares secret sauce (if necessary) during due diligence
3) Big Corp makes a super low offer
4) Big Corp puts 2 engineers on it to copy the result
5) Big Corp is now Bigger Corp, and single dev is still single.
Yes, this is exactly how it will go down I think. Somewhere I read a great analysis which said this was the opposite of "disruptive innovation", but instead this type of innovation will primarily reinforce the positions of the incumbents.
I've met people who applied to it at different times. None have got the key yet, including me. I don't think they are handing out many api keys. They probably had some sort of limit in mind when they launched and they probably reached it already. Unless you're a company or individual they are particularly interested in, I think the chances of getting it in the near future are quite slim :(
Looks very promising (although I have no API key)!
Feature request:
I got surprised when I was asked to sign in. Introduce an anonymous mode where "Sign in with Google" is not enforced. History could be exported locally and imported from the file system.
Does anyone know if they'll ever release the models for GPT-3 so we can train/re-train then ourselves? Or is GPT-3 so general that it doesn't need retraining?
I don’t know if they will release the models, but are you sure you can train a 170 billion parameter model? Last I heard it’s around 500GB, which would require serious infrastructure.
What's interesting with machine learning is that in a few years time algorithms get efficient enough to train the same quality models on commodity hardware. At the same time organizations are always a few years ahead :(
If GPT-3 was trained on the Internet and even if some information is publicly available, then solution described might be somehow legally protected.
Does GPT-3 recognises that?
Can I request a share in selling of GPT-3 license if I prove that uses my solutions?
I've been there myself tbh. When you have worked a lot on something, it's often hard to take a step back and specify all of the basic things that people first exposed to your invention don't know about.